Installation of NVIDIA Agent Toolkit requires Python 3.11 or 3.12 and the uv package manager. Start by cloning the repository: git clone -b main https://github.com/NVIDIA/NeMo-Agent-Toolkit.git nemo-agent-toolkit. Navigate to the directory and initialize submodules with git submodule update --init --recursive. Download datasets with git lfs install/fetch/pull to ensure all training data is available locally.
Create a Python virtual environment using uv sync --all-groups --all-extras, which installs the toolkit and all optional framework integrations (LangChain, LlamaIndex, CrewAI, etc.). Verify installation by running nat --version to check the version and access help documentation. For framework-specific installations, use extras: nvidia-nat[langchain] for LangChain, nvidia-nat[llamaindex] for LlamaIndex, etc.
Avoid Conda environments, which can cause component resolution conflicts. Use only vanilla Python virtual environments via python -m venv or uv venv. The official NVIDIA NeMo Agent Toolkit documentation includes platform-specific instructions and troubleshooting. For managed deployments, Zilliz Cloud provides pre-integrated environments where agents are provisioned alongside vector database infrastructure, eliminating manual setup.